fix: delegate offline mode control from container template to train.py#16
Merged
Conversation
The container SLURM template was hardcoding HF_HUB_OFFLINE=1 etc. regardless of config.offline, so offline: false had no effect inside the container and jobs would stall trying to reach the Hub at runtime. Remove the three hardcoded offline flags from job_trl_container.sh.jinja and let train.py own this: the existing if config.offline block sets them to 1, and a new else block explicitly sets them to 0 to clear any value inherited from the container environment.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The containerised TRL SLURM template (
job_trl_container.sh.jinja) was hardcodingHF_HUB_OFFLINE=1,HF_DATASETS_OFFLINE=1, andTRANSFORMERS_OFFLINE=1inside the container regardless ofconfig.offline. This meantoffline: falsehad no effect at runtime — the container always ran in offline mode and would stall if models weren't already cached.This PR removes those hardcoded flags from the template and delegates ownership to
train.py, which already readsconfig.offlineto set these vars. Anelsebranch is added to explicitly zero out the flags whenoffline: false, guarding against any stale values inherited from the container environment.Type of change